The effectiveness of two selection methods for the improvement of grain sorghum (Sorghum bicolor (L.) Moench) populations was compared in two experiments. IAP1R(M) was developed by using mass selection for individual-panicle grain weight, while IAP4R(S1) was developed from the same base population by using replicated yield tests of S1 families;Results from Experiment I indicated that the selection methods significantly (P 0.05) between selection schemes for the improvement of grain yield of the C2 and C3 populations. Gain per cycle was estimated at 0.117 Mg/ha (2.06%/cycle) for mass selection and 0.170 Mg/ha (2.99%/cycle) for S1 testing;Compared to the C0 population, all advanced-cycle-composites of IAP4R(S1) averaged significantly (P 0.05) between the C3 populations in genetic variance among S1 families for grain yield, panicles/plant, and plant height (Experiment II). Estimates of genetic variance for seeds/panicle and days to midbloom were larger in IAP4R(S1)C3, but smaller for 100-seed weight, compared to IAP1R(M)C3. Strong negative correlations of seeds/panicle with 100-seed weight indicated that simultaneous selection for more seeds/panicle and large seed would be difficult.